diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_full_8p.sh b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_full_8p.sh
index e23728d377b1bde542b4110901b1e8114db47b95..74ea8a80f5f77678c1134c12aa218072ddba0057 100644
--- a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_full_8p.sh
+++ b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_full_8p.sh
@@ -5,8 +5,8 @@ cur_path=`pwd`
 
 #集合通信参数,不需要修改
 #保证rank table file 文件rank_table_8p.json存放在和test同级的configs目录下
-export RANK_SIZE=8
-export RANK_TABLE_FILE=${cur_path}/../8p.json
+export RANK_SIZES=8
+#export RANK_TABLE_FILE=${cur_path}/../8p.json
 export JOB_ID=10087
 RANK_ID_START=0
 
@@ -89,9 +89,19 @@ do
     elif [[ $para == --bind_core* ]]; then
         bind_core=`echo ${para#*=}`
         name_bind="_bindcore"
+    elif [[ $para == --one_node_ip* ]];then
+        one_node_ip=`echo ${para#*=}`
     fi
 done
 
+#8p训练必须参数（本机IP）
+one_node_ip=$one_node_ip
+#新增适配集群环境变量
+export CM_CHIEF_IP=${one_node_ip}   #主节点ip，所有服务器一致
+export CM_CHIEF_PORT=29688          #通信端口，所有服务器一致
+export CM_CHIEF_DEVICE=0            #配置为0，配置主卡，类似于主节点，所有服务器一致
+export CM_WORKER_SIZE=8             #卡数，单机为8，所有服务器一致
+export CM_WORKER_IP=${one_node_ip}  #当前服务器ip，不同环境ip不同
 #校验是否传入data_path,不需要修改
 if [[ $data_path == "" ]];then
     echo "[Error] para \"data_path\" must be confing"
@@ -108,15 +118,17 @@ fi
 #训练开始时间，不需要修改
 start_time=$(date +%s)
 
+sed -i 's/RANK_SIZE/RANK_SIZES/g' ../modelarts/start.py ../efficientnet/main_npu.py
+sed -i 's/RANK_ID/RANK_IDS/g' ../modelarts/start.py
 #进入训练脚本目录，需要模型审视修改
 cd $cur_path/../
-for((RANK_ID=$RANK_ID_START;RANK_ID<$((RANK_SIZE+RANK_ID_START));RANK_ID++));
+for((RANK_IDS=$RANK_ID_START;RANK_IDS<$((RANK_SIZES+RANK_ID_START));RANK_IDS++));
 do
     #设置环境变量，不需要修改
-    echo "Device ID: $RANK_ID"
-    export RANK_ID=$RANK_ID
-    export ASCEND_DEVICE_ID=$RANK_ID
-    ASCEND_DEVICE_ID=$RANK_ID
+    echo "Device ID: $RANK_IDS"
+    export RANK_IDS=$RANK_IDS
+    export ASCEND_DEVICE_ID=$RANK_IDS
+    ASCEND_DEVICE_ID=$RANK_IDS
     
     #创建DeviceID输出目录，不需要修改
     if [ -d ${cur_path}/output/${ASCEND_DEVICE_ID} ];then
@@ -128,9 +140,9 @@ do
     
      # 绑核，不需要的绑核的模型删除，需要模型审视修改
     corenum=`cat /proc/cpuinfo |grep "processor"|wc -l`
-    let a=RANK_ID*${corenum}/${RANK_SIZE}
-    let b=RANK_ID+1
-    let c=b*${corenum}/${RANK_SIZE}-1
+    let a=RANK_IDS*${corenum}/${RANK_SIZES}
+    let b=RANK_IDS+1
+    let c=b*${corenum}/${RANK_SIZES}-1
 
     #执行训练脚本，以下传参不需要修改，其他需要模型审视修改
     #--data_dir, --model_dir, --precision_mode, --over_dump, --over_dump_path，--data_dump_flag，--data_dump_step，--data_dump_path，--profiling，--profiling_dump_path
@@ -153,7 +165,8 @@ wait
 #训练结束时间，不需要修改
 end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
-
+sed -i 's/RANK_SIZES/RANK_SIZE/g' modelarts/start.py efficientnet/main_npu.py
+sed -i 's/RANK_IDS/RANK_ID/g' modelarts/start.py
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
@@ -171,13 +184,13 @@ echo "E2E Training Duration sec : $e2e_time"
 #训练用例信息，不需要修改
 BatchSize=${batch_size}
 DeviceType=`uname -m`
-CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZE}'p'_'acc'
+CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZES}'p'_'acc'
 
 ##获取性能数据
 #吞吐量，不需要修改
 ActualFPS=${FPS}
 #单迭代训练时长，不需要修改
-TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZE}'*1000/'${FPS}'}'`
+TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZES}'*1000/'${FPS}'}'`
 
 #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中，需要根据模型审视
 grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
@@ -187,7 +200,7 @@ ActualLoss=`awk 'END {print}' $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseNam
 
 #关键信息打印到${CaseName}.log中，不需要修改
 echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZES}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "BatchSize = ${BatchSize}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "DeviceType = ${DeviceType}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
@@ -195,4 +208,4 @@ echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName
 echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
\ No newline at end of file
diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_16p.sh b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_16p.sh
index 24aaf972b09aac08f47ae94f05381403e76f58b0..7bce07eb4eadea2a90fcc418fc55bc9a7511eac6 100644
--- a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_16p.sh
+++ b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_16p.sh
@@ -3,17 +3,13 @@
 cur_path=`pwd`
 
 #集合通信参数,不需要修改
-#保证rank table file 文件rank_table_8p.json存放在和test同级的configs目录下
-#export RANK_SIZE=8
-#export RANK_TABLE_FILE=${cur_path}/../8p.json
-#export JOB_ID=10087
 RANK_ID_START=0
 
 # 数据集路径,保持为空,不需要修改
 data_path="/npu/traindata/imagenet_TF"
 
 #设置默认日志级别,不需要修改
-export ASCEND_GLOBAL_LOG_LEVEL=3
+#export ASCEND_GLOBAL_LOG_LEVEL_ETP=3
 
 #基础参数 需要模型审视修改
 #网络名称，同目录名称
@@ -93,9 +89,27 @@ do
         server_index=`echo ${para#*=}`
     elif [[ $para == --conf_path* ]];then
         conf_path=`echo ${para#*=}`
+    elif [[ $para == --fix_node_ip* ]];then
+	    fix_node_ip=`echo ${para#*=}`
+    elif [[ $para == --one_node_ip* ]];then
+        one_node_ip=`echo ${para#*=}`
     fi
 done
 
+if [[ $conf_path == "" ]];then
+    fix_node_ip=$fix_node_ip
+    one_node_ip=$one_node_ip
+else
+    one_node_ip=`find $conf_path -name "server_*_0.info"|awk -F "server_" '{print $2}'|awk -F "_" '{print $1}'`
+fi
+
+#新增适配集群环境变量
+export CM_CHIEF_IP=${one_node_ip}   #主节点ip，所有服务器一致
+export CM_CHIEF_PORT=29688          #通信端口，所有服务器一致
+export CM_CHIEF_DEVICE=0            #配置为0，配置主卡，类似于主节点，所有服务器一致
+export CM_WORKER_SIZE=16            #卡数，单机为8，多机为8n,所有服务器一致
+export CM_WORKER_IP=${fix_node_ip}  #当前服务器ip，不同环境ip不同
+
 #校验是否传入data_path,不需要修改
 if [[ $data_path == "" ]];then
     echo "[Error] para \"data_path\" must be confing"
@@ -111,27 +125,29 @@ fi
 
 #训练开始时间，不需要修改
 start_time=$(date +%s)
-export RANK_SIZE=16
+export RANK_SIZES=16
 rank_size=8
-    
-if [[ $conf_path != "" ]];then
-    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZE/rank_size)) --conf_path=$conf_path
-fi
 
-wait
-export RANK_TABLE_FILE=${cur_path}/rank_table.json
+#if [[ $conf_path != "" ]];then
+#    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZE/rank_size)) --conf_path=$conf_path
+#fi
+#
+#wait
+#export RANK_TABLE_FILE=${cur_path}/rank_table.json
 export JOB_ID=10087
 export DEVICE_INDEX=0
 
+sed -i 's/RANK_SIZE/RANK_SIZES/g' ../modelarts/start.py ../efficientnet/main_npu.py
+sed -i 's/RANK_ID/RANK_IDS/g' ../modelarts/start.py
 #进入训练脚本目录，需要模型审视修改
 cd $cur_path/../
-for((RANK_ID=$((rank_size*server_index));RANK_ID<$((((server_index+1))*rank_size));RANK_ID++));
+for((RANK_IDS=$((rank_size*server_index));RANK_IDS<$((((server_index+1))*rank_size));RANK_IDS++));
 do
     #设置环境变量，不需要修改
-    echo "Device ID: $RANK_ID"
-    export RANK_ID=$RANK_ID
-    export ASCEND_DEVICE_ID=`expr ${RANK_ID} - $((rank_size*server_index))`
-    ASCEND_DEVICE_ID=`expr ${RANK_ID} - $((rank_size*server_index))` 
+    echo "Device ID: $RANK_IDS"
+    export RANK_IDS=$RANK_IDS
+    export ASCEND_DEVICE_ID=`expr ${RANK_IDS} - $((rank_size*server_index))`
+    ASCEND_DEVICE_ID=`expr ${RANK_IDS} - $((rank_size*server_index))`
     #创建DeviceID输出目录，不需要修改
     if [ -d ${cur_path}/output/${ASCEND_DEVICE_ID} ];then
         rm -rf ${cur_path}/output/${ASCEND_DEVICE_ID}
@@ -142,9 +158,9 @@ do
     
      # 绑核，不需要的绑核的模型删除，需要模型审视修改
     corenum=`cat /proc/cpuinfo |grep "processor"|wc -l`
-    let a=RANK_ID*${corenum}/${RANK_SIZE}
-    let b=RANK_ID+1
-    let c=b*${corenum}/${RANK_SIZE}-1
+    let a=RANK_IDS*${corenum}/${RANK_SIZES}
+    let b=RANK_IDS+1
+    let c=b*${corenum}/${RANK_SIZES}-1
 
     #执行训练脚本，以下传参不需要修改，其他需要模型审视修改
     #--data_dir, --model_dir, --precision_mode, --over_dump, --over_dump_path，--data_dump_flag，--data_dump_step，--data_dump_path，--profiling，--profiling_dump_path
@@ -168,7 +184,8 @@ wait
 #训练结束时间，不需要修改
 end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
-
+sed -i 's/RANK_SIZES/RANK_SIZE/g' modelarts/start.py efficientnet/main_npu.py
+sed -i 's/RANK_IDS/RANK_ID/g' modelarts/start.py
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
@@ -187,13 +204,13 @@ echo "E2E Training Duration sec : $e2e_time"
 #训练用例信息，不需要修改
 BatchSize=${batch_size}
 DeviceType=`uname -m`
-CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'
+CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZES}'p'_'perf'
 
 ##获取性能数据
 #吞吐量，不需要修改
 ActualFPS=${FPS}
 #单迭代训练时长，不需要修改
-TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZE}'*1000/'${FPS}'}'`
+TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZES}'*1000/'${FPS}'}'`
 
 #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中，需要根据模型审视
 grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
@@ -203,7 +220,7 @@ ActualLoss=`awk 'END {print}' $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseNam
 
 #关键信息打印到${CaseName}.log中，不需要修改
 echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZES}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "BatchSize = ${BatchSize}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "DeviceType = ${DeviceType}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
@@ -211,4 +228,4 @@ echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName
 echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 #echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
\ No newline at end of file
diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_32p.sh b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_32p.sh
index 50478ea1d1e7c84b84e2a11a9f47fe6c0a7a8037..a3824dc961d3ef75223e2b8ea62dbcd51cc2c634 100644
--- a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_32p.sh
+++ b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_32p.sh
@@ -3,17 +3,13 @@
 cur_path=`pwd`
 export ASCEND_HOST_LOG_FILE_NUM=1000 
 #集合通信参数,不需要修改
-#保证rank table file 文件rank_table_8p.json存放在和test同级的configs目录下
-#export RANK_SIZE=8
-#export RANK_TABLE_FILE=${cur_path}/../8p.json
-#export JOB_ID=10087
 RANK_ID_START=0
 
 # 数据集路径,保持为空,不需要修改
 data_path="/npu/traindata/imagenet_TF"
 
 #设置默认日志级别,不需要修改
-export ASCEND_GLOBAL_LOG_LEVEL_ETP=3
+#export ASCEND_GLOBAL_LOG_LEVEL_ETP_ETP=3
 
 #基础参数 需要模型审视修改
 #网络名称，同目录名称
@@ -93,9 +89,26 @@ do
         server_index=`echo ${para#*=}`
     elif [[ $para == --conf_path* ]];then
         conf_path=`echo ${para#*=}`
+    elif [[ $para == --fix_node_ip* ]];then
+	    fix_node_ip=`echo ${para#*=}`
+    elif [[ $para == --one_node_ip* ]];then
+        one_node_ip=`echo ${para#*=}`
     fi
 done
 
+if [[ $conf_path == "" ]];then
+    fix_node_ip=$fix_node_ip
+    one_node_ip=$one_node_ip
+else
+    one_node_ip=`find $conf_path -name "server_*_0.info"|awk -F "server_" '{print $2}'|awk -F "_" '{print $1}'`
+fi
+
+#新增适配集群环境变量
+export CM_CHIEF_IP=${one_node_ip}   #主节点ip，所有服务器一致
+export CM_CHIEF_PORT=29688          #通信端口，所有服务器一致
+export CM_CHIEF_DEVICE=0            #配置为0，配置主卡，类似于主节点，所有服务器一致
+export CM_WORKER_SIZE=32            #卡数，单机为8，多机为8n,所有服务器一致
+export CM_WORKER_IP=${fix_node_ip}  #当前服务器ip，不同环境ip不同
 #校验是否传入data_path,不需要修改
 if [[ $data_path == "" ]];then
     echo "[Error] para \"data_path\" must be confing"
@@ -111,18 +124,20 @@ fi
 
 #训练开始时间，不需要修改
 start_time=$(date +%s)
-export RANK_SIZE=32
+export RANK_SIZES=32
 rank_size=8
 
-if [[ $conf_path != "" ]];then
-    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZE/rank_size)) --conf_path=$conf_path
-fi
-
-wait
-export RANK_TABLE_FILE=${cur_path}/rank_table.json
+#if [[ $conf_path != "" ]];then
+#    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZES/rank_size)) --conf_path=$conf_path
+#fi
+#
+#wait
+#export RANK_TABLE_FILE=${cur_path}/rank_table.json
 export JOB_ID=10087
 export DEVICE_INDEX=0
 
+sed -i 's/RANK_SIZE/RANK_SIZES/g' ../modelarts/start.py ../efficientnet/main_npu.py
+sed -i 's/RANK_ID/RANK_IDS/g' ../modelarts/start.py
 #进入训练脚本目录，需要模型审视修改
 cd $cur_path/../
 for((RANK_ID=$((rank_size*server_index));RANK_ID<$((((server_index+1))*rank_size));RANK_ID++));
@@ -142,9 +157,9 @@ do
     
      # 绑核，不需要的绑核的模型删除，需要模型审视修改
     corenum=`cat /proc/cpuinfo |grep "processor"|wc -l`
-    let a=RANK_ID*${corenum}/${RANK_SIZE}
+    let a=RANK_ID*${corenum}/${RANK_SIZES}
     let b=RANK_ID+1
-    let c=b*${corenum}/${RANK_SIZE}-1
+    let c=b*${corenum}/${RANK_SIZES}-1
 
     #执行训练脚本，以下传参不需要修改，其他需要模型审视修改
     #--data_dir, --model_dir, --precision_mode, --over_dump, --over_dump_path，--data_dump_flag，--data_dump_step，--data_dump_path，--profiling，--profiling_dump_path
@@ -168,7 +183,8 @@ wait
 #训练结束时间，不需要修改
 end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
-
+sed -i 's/RANK_SIZES/RANK_SIZE/g' modelarts/start.py efficientnet/main_npu.py
+sed -i 's/RANK_IDS/RANK_ID/g' modelarts/start.py
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
@@ -187,13 +203,13 @@ echo "E2E Training Duration sec : $e2e_time"
 #训练用例信息，不需要修改
 BatchSize=${batch_size}
 DeviceType=`uname -m`
-CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'
+CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZES}'p'_'perf'
 
 ##获取性能数据
 #吞吐量，不需要修改
 ActualFPS=${FPS}
 #单迭代训练时长，不需要修改
-TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZE}'*1000/'${FPS}'}'`
+TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZES}'*1000/'${FPS}'}'`
 
 #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中，需要根据模型审视
 grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' > $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
@@ -203,7 +219,7 @@ ActualLoss=`awk 'END {print}' $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseNam
 
 #关键信息打印到${CaseName}.log中，不需要修改
 echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZES}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "BatchSize = ${BatchSize}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "DeviceType = ${DeviceType}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
@@ -211,5 +227,4 @@ echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName
 echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 #echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-
+echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
\ No newline at end of file
diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_64p.sh b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_64p.sh
index acbce4c11ba8789051996626d00680b9bff030fb..ce3e639740225912ce5aa14b1df92150b1a8f735 100644
--- a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_64p.sh
+++ b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_64p.sh
@@ -3,17 +3,13 @@
 cur_path=`pwd`
 
 #集合通信参数,不需要修改
-#保证rank table file 文件rank_table_8p.json存放在和test同级的configs目录下
-#export RANK_SIZE=8
-#export RANK_TABLE_FILE=${cur_path}/../8p.json
-#export JOB_ID=10087
 RANK_ID_START=0
 
 # 数据集路径,保持为空,不需要修改
 data_path="/npu/traindata/imagenet_TF"
 
 #设置默认日志级别,不需要修改
-export ASCEND_GLOBAL_LOG_LEVEL_ETP=3
+#export ASCEND_GLOBAL_LOG_LEVEL_ETP_ETP=3
 
 #基础参数 需要模型审视修改
 #网络名称，同目录名称
@@ -93,9 +89,27 @@ do
         server_index=`echo ${para#*=}`
     elif [[ $para == --conf_path* ]];then
         conf_path=`echo ${para#*=}`
+    elif [[ $para == --fix_node_ip* ]];then
+	    fix_node_ip=`echo ${para#*=}`
+    elif [[ $para == --one_node_ip* ]];then
+        one_node_ip=`echo ${para#*=}`
     fi
 done
 
+if [[ $conf_path == "" ]];then
+    fix_node_ip=$fix_node_ip
+    one_node_ip=$one_node_ip
+else
+    one_node_ip=`find $conf_path -name "server_*_0.info"|awk -F "server_" '{print $2}'|awk -F "_" '{print $1}'`
+fi
+
+#新增适配集群环境变量
+export CM_CHIEF_IP=${one_node_ip}   #主节点ip，所有服务器一致
+export CM_CHIEF_PORT=29688          #通信端口，所有服务器一致
+export CM_CHIEF_DEVICE=0            #配置为0，配置主卡，类似于主节点，所有服务器一致
+export CM_WORKER_SIZE=64            #卡数，单机为8，多机为8n,所有服务器一致
+export CM_WORKER_IP=${fix_node_ip}  #当前服务器ip，不同环境ip不同
+
 #校验是否传入data_path,不需要修改
 if [[ $data_path == "" ]];then
     echo "[Error] para \"data_path\" must be confing"
@@ -111,27 +125,29 @@ fi
 
 #训练开始时间，不需要修改
 start_time=$(date +%s)
-export RANK_SIZE=64
+export RANK_SIZES=64
 rank_size=8
 
-if [[ $conf_path != "" ]];then
-    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZE/rank_size)) --conf_path=$conf_path
-fi
-
-wait
-export RANK_TABLE_FILE=${cur_path}/rank_table.json
+#if [[ $conf_path != "" ]];then
+#    nohup python3 set_ranktable.py --npu_nums=$((RANK_SIZE/rank_size)) --conf_path=$conf_path
+#fi
+#
+#wait
+#export RANK_TABLE_FILE=${cur_path}/rank_table.json
 export JOB_ID=10087
 export DEVICE_INDEX=0
 
+sed -i 's/RANK_SIZE/RANK_SIZES/g' ../modelarts/start.py ../efficientnet/main_npu.py
+sed -i 's/RANK_ID/RANK_IDS/g' ../modelarts/start.py
 #进入训练脚本目录，需要模型审视修改
 cd $cur_path/../
-for((RANK_ID=$((rank_size*server_index));RANK_ID<$((((server_index+1))*rank_size));RANK_ID++));
+for((RANK_IDS=$((rank_size*server_index));RANK_IDS<$((((server_index+1))*rank_size));RANK_IDS++));
 do
     #设置环境变量，不需要修改
-    echo "Device ID: $RANK_ID"
-    export RANK_ID=$RANK_ID
-    export ASCEND_DEVICE_ID=`expr ${RANK_ID} - $((rank_size*server_index))`
-    ASCEND_DEVICE_ID=`expr ${RANK_ID} - $((rank_size*server_index))` 
+    echo "Device ID: $RANK_IDS"
+    export RANK_IDS=$RANK_IDS
+    export ASCEND_DEVICE_ID=`expr ${RANK_IDS} - $((rank_size*server_index))`
+    ASCEND_DEVICE_ID=`expr ${RANK_IDS} - $((rank_size*server_index))`
     #创建DeviceID输出目录，不需要修改
     if [ -d ${cur_path}/output/${ASCEND_DEVICE_ID} ];then
         rm -rf ${cur_path}/output/${ASCEND_DEVICE_ID}
@@ -142,9 +158,9 @@ do
     
      # 绑核，不需要的绑核的模型删除，需要模型审视修改
     corenum=`cat /proc/cpuinfo |grep "processor"|wc -l`
-    let a=RANK_ID*${corenum}/${RANK_SIZE}
-    let b=RANK_ID+1
-    let c=b*${corenum}/${RANK_SIZE}-1
+    let a=RANK_IDS*${corenum}/${RANK_SIZES}
+    let b=RANK_IDS+1
+    let c=b*${corenum}/${RANK_SIZES}-1
 
     #执行训练脚本，以下传参不需要修改，其他需要模型审视修改
     #--data_dir, --model_dir, --precision_mode, --over_dump, --over_dump_path，--data_dump_flag，--data_dump_step，--data_dump_path，--profiling，--profiling_dump_path
@@ -168,7 +184,8 @@ wait
 #训练结束时间，不需要修改
 end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
-
+sed -i 's/RANK_SIZES/RANK_SIZE/g' modelarts/start.py efficientnet/main_npu.py
+sed -i 's/RANK_IDS/RANK_ID/g' modelarts/start.py
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
@@ -187,13 +204,13 @@ echo "E2E Training Duration sec : $e2e_time"
 #训练用例信息，不需要修改
 BatchSize=${batch_size}
 DeviceType=`uname -m`
-CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'
+CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZES}'p'_'perf'
 
 ##获取性能数据
 #吞吐量，不需要修改
 ActualFPS=${FPS}
 #单迭代训练时长，不需要修改
-TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZE}'*1000/'${FPS}'}'`
+TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZES}'*1000/'${FPS}'}'`
 
 #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中，需要根据模型审视
 grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
@@ -203,7 +220,7 @@ ActualLoss=`awk 'END {print}' $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseNam
 
 #关键信息打印到${CaseName}.log中，不需要修改
 echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZES}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "BatchSize = ${BatchSize}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "DeviceType = ${DeviceType}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
@@ -211,5 +228,4 @@ echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName
 echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 #echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-
+echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
\ No newline at end of file
diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_8p.sh b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_8p.sh
index 29d51a154d6f660accd2dce4e3aab6e02d6d56d9..b9038f8e13f56f9edc0746aab99dea4da77ae07f 100644
--- a/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_8p.sh
+++ b/TensorFlow/built-in/cv/image_classification/EfficientNet_B0_ID0009_for_TensorFlow/test/train_performance_8p.sh
@@ -4,8 +4,8 @@ cur_path=`pwd`
 
 #集合通信参数,不需要修改
 #保证rank table file 文件rank_table_8p.json存放在和test同级的configs目录下
-export RANK_SIZE=8
-export RANK_TABLE_FILE=${cur_path}/../8p.json
+export RANK_SIZES=8
+#export RANK_TABLE_FILE=${cur_path}/../8p.json
 export JOB_ID=10087
 RANK_ID_START=0
 
@@ -13,7 +13,7 @@ RANK_ID_START=0
 data_path="/npu/traindata/imagenet_TF"
 
 #设置默认日志级别,不需要修改
-export ASCEND_GLOBAL_LOG_LEVEL=3
+#export ASCEND_GLOBAL_LOG_LEVEL_ETP=3
 
 #基础参数 需要模型审视修改
 #网络名称，同目录名称
@@ -89,9 +89,19 @@ do
     elif [[ $para == --bind_core* ]]; then
         bind_core=`echo ${para#*=}`
         name_bind="_bindcore"
+    elif [[ $para == --one_node_ip* ]];then
+        one_node_ip=`echo ${para#*=}`
     fi
 done
 
+#8p训练必须参数（本机IP）
+one_node_ip=$one_node_ip
+#新增适配集群环境变量
+export CM_CHIEF_IP=${one_node_ip}   #主节点ip，所有服务器一致
+export CM_CHIEF_PORT=29688          #通信端口，所有服务器一致
+export CM_CHIEF_DEVICE=0            #配置为0，配置主卡，类似于主节点，所有服务器一致
+export CM_WORKER_SIZE=8             #卡数，单机为8，所有服务器一致
+export CM_WORKER_IP=${one_node_ip}  #当前服务器ip，不同环境ip不同
 #校验是否传入data_path,不需要修改
 if [[ $data_path == "" ]];then
     echo "[Error] para \"data_path\" must be confing"
@@ -108,15 +118,17 @@ fi
 #训练开始时间，不需要修改
 start_time=$(date +%s)
 
+sed -i 's/RANK_SIZE/RANK_SIZES/g' ../modelarts/start.py ../efficientnet/main_npu.py
+sed -i 's/RANK_ID/RANK_IDS/g' ../modelarts/start.py
 #进入训练脚本目录，需要模型审视修改
 cd $cur_path/../
-for((RANK_ID=$RANK_ID_START;RANK_ID<$((RANK_SIZE+RANK_ID_START));RANK_ID++));
+for((RANK_IDS=$RANK_ID_START;RANK_IDS<$((RANK_SIZES+RANK_ID_START));RANK_IDS++));
 do
     #设置环境变量，不需要修改
-    echo "Device ID: $RANK_ID"
-    export RANK_ID=$RANK_ID
-    export ASCEND_DEVICE_ID=$RANK_ID
-    ASCEND_DEVICE_ID=$RANK_ID
+    echo "Device ID: $RANK_IDS"
+    export RANK_IDS=$RANK_IDS
+    export ASCEND_DEVICE_ID=$RANK_IDS
+    ASCEND_DEVICE_ID=$RANK_IDS
     
     #创建DeviceID输出目录，不需要修改
     if [ -d ${cur_path}/output/${ASCEND_DEVICE_ID} ];then
@@ -128,9 +140,9 @@ do
     
      # 绑核，不需要的绑核的模型删除，需要模型审视修改
     corenum=`cat /proc/cpuinfo |grep "processor"|wc -l`
-    let a=RANK_ID*${corenum}/${RANK_SIZE}
-    let b=RANK_ID+1
-    let c=b*${corenum}/${RANK_SIZE}-1
+    let a=RANK_IDS*${corenum}/${RANK_SIZES}
+    let b=RANK_IDS+1
+    let c=b*${corenum}/${RANK_SIZES}-1
 
     #执行训练脚本，以下传参不需要修改，其他需要模型审视修改
     #--data_dir, --model_dir, --precision_mode, --over_dump, --over_dump_path，--data_dump_flag，--data_dump_step，--data_dump_path，--profiling，--profiling_dump_path
@@ -154,7 +166,8 @@ wait
 #训练结束时间，不需要修改
 end_time=$(date +%s)
 e2e_time=$(( $end_time - $start_time ))
-
+sed -i 's/RANK_SIZES/RANK_SIZE/g' modelarts/start.py efficientnet/main_npu.py
+sed -i 's/RANK_IDS/RANK_ID/g' modelarts/start.py
 #结果打印，不需要修改
 echo "------------------ Final result ------------------"
 #输出性能FPS，需要模型审视修改
@@ -173,23 +186,23 @@ echo "E2E Training Duration sec : $e2e_time"
 #训练用例信息，不需要修改
 BatchSize=${batch_size}
 DeviceType=`uname -m`
-CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'
+CaseName=${Network}${name_bind}_bs${BatchSize}_${RANK_SIZES}'p'_'perf'
 
 ##获取性能数据
 #吞吐量，不需要修改
 ActualFPS=${FPS}
 #单迭代训练时长，不需要修改
-TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZE}'*1000/'${FPS}'}'`
+TrainingTime=`awk 'BEGIN{printf "%.2f\n",'${BatchSize}'*'${RANK_SIZES}'*1000/'${FPS}'}'`
 
 #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中，需要根据模型审视
-grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
+grep 'logger.py:54' $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log|awk '{print $8}' |awk -F ":" '{print $2}' > $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt
 
 #最后一个迭代loss值，不需要修改
 ActualLoss=`awk 'END {print}' $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt`
 
 #关键信息打印到${CaseName}.log中，不需要修改
 echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
-echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZES}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "BatchSize = ${BatchSize}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "DeviceType = ${DeviceType}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
 echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log